AIIC AI Intelligence Centre

SOURCE-LINKED INTELLIGENCE

ESTS at WMT26: Routing-Informed Expert Pruning for Model Compression

arXiv · AI, language, vision and robotics · article · Sep 11, 2026 · UTC

We describe six submissions under the team name ESTS to the unconstrained WMT26 Model Compression Shared Task for English--Simplified Chinese and English--Egyptian Arabic. We submit three compression operating points per translation direction, all derived from GPT-OSS-20B. We use task-specific routing mass to rank experts and cross-lingual routing divergence to allocate retained capacity across layers, then physically remove low-importance experts. The resulting specialists are recovery-tuned on GPT-5.1-generated synthetic translation data and further compressed by applying MXFP4 quantization

Read original source ↗ Open in workspace

recordType
paper
region
Global

Evidence & attribution

First collected: 2026-09-20T18:42:18.733Z. This is not the publication date.